Development and evaluation of a proxy for baseline ECOG PS in advanced non-small cell lung cancer (aNSCLC), bladder cancer (aBCa), and melanoma (aMM): An electronic health records (EHR) study.

Abstract

e18194 Background: Performance status measured by ECOG PS (ECOG) predicts therapy choice, and overall survival (OS), but often not reported in routine care. EHR data include rich clinical information which Methods: In this retrospective analysis using Flatiron Health EHR-derived database we selected patients diagnosed from 2011-2018 with aNSCLC, aBCa, and aMM and ECOG reported between diagnosis and 1) therapy start, or 2) within 90 days. We used multivariable logistic regressions in a 75% development sample to model associations between selected sociodemographic, clinical and laboratory measures and poor ECOG (2+ vs 0-1) for each cancer. Predictive accuracy was assessed in 25% validation datasets. Kaplan-Meier analysis tested the associations between observed and predicted poor ECOG with OS. Results: The non-missing ECOG samples were 20,697 aNSCLC, 2627 aBCa, and 2558 aMM (out of 50,876, 6955, and 7917 respectively). The proportion of patients with ECOG 2+ ranged from 28.5% for aNSCLC to 15% for aMM. Higher number of comorbid conditions, older age, lowest BMI, higher number of metastases, abnormal albumin and hemoglobin and public insurance were associated with poor ECOG status. Predictive accuracy for poor baseline ECOG was 71.8% for NSLCL, 73.8% for aBca, and 85.4% for aMM). There was a consistent association between OS and observed and predicted baseline ECOG (Table 1). Conclusions: In a large cohort of cancer patients, routinely collected EHR data were used to predict baseline ECOG PS status. Complete ECOG status results may enhance evaluation of cancer treatment patterns and outcomes in EHR data. [Table: see text]